The small signal stability margin prediction model for grid-connected hybrid renewable power station based on the GCN-BiLSTM

With the large-scale access of high-proportion renewable power and power electronic equipment, power system obviously shows a trend of decreasing inertia and weakening system strength, and the stability problem is becoming more and more serious. In addition to system operation status data, the topology in the hybrid renewable power station also affects the small signal stability of the grid-connected system. However, existing data driven prediction model of grid-connected stability margin of renewable power stations cannot characterize the effect of its topology. Therefore, if the topology in the station changes, its prediction accuracy will be difficult to guarantee. Considering the influence of topology, the grid-connected stability margin prediction model of grid-forming/grid-following hybrid renewable power station based on Graph Convolution Networks (GCN) and Bi-directional long short-term memory (BiLSTM) , namely GCN-BiLSTM, is proposed. GCN is used to realize the perception of the topology structure in the renewable power station, so as to be adapt to the change of its topology and improve the topology generalization ability of the model. BiLSTM is used to deeply fit the nonlinear relationship between renewable power station topology, system operation status and stability margin, so as to further improve the prediction accuracy of GCN-BiLSTM model. The information of power grid strength, the configuration ratio and distribution location of grid-forming/grid-following power generation equipment, type of energy storage device and topology of the station are input into the GCN-BiLSTM prediction model. And the prediction results of small signal stability margin in low frequency band and sub/super-synchronous band are output online. Taking a demo grid-forming/grid-following hybrid renewable power station with energy storage device as an example, the feasibility and effectiveness of the proposed grid-connected renewable power station stability margin prediction model based on the GCN-BiLSTM are verified. The experimental results show that compared with the prediction model based on the BiLSTM, GCN-BiLSTM has better topological generalization ability, and higher prediction accuracy.